Voon, Cherng Jyh (2020) An Educational Android App For Identifying Animals In Zoo. Final Year Project, UTAR.
Abstract
Zoo has been one of the favourite learning spots for children when it comes to animal study. In the outdoor learning session, teachers will often find it difficult to tutor their students while ensuring their safety. The advent of modern technology aided teachers in educating students without their presence. The project explored the possibility of using smartphones as a medium to help kids to learn about animals in zoos, even without the guidance of teachers. The objective of this project is to build an Android mobile application for kids that automatically recognizes images of an animal using Convolutional Neural Network and Transfer Learning. Phased development methodology was used for application development. The developed application consists of three major modules: animal recognition module that is able to recognize 18 different animals, gamification module that attracts the interest of kids and lastly quiz module that broadens the kid’s knowledge on animals. Apart from that, experiments were performed on three pretrained models, which is MobileNet, EffNet and NASNet to determine the most suitable model to be used for Transfer Learning. The models were analyzed from the aspect of accuracy, output model size, speed and power consumption. MobileNet was observed to be the least power-hungry (38.2 million FLOPs), fastest (0.018 Million parameters) and smallest in size (12.859 MB) among the three. Considering the time spent in zoo, Mobilenet was integrated into the system for image recognition. The mobile application was successfully designed, developed and tested throughout the project. Three pretrained models were compared and MobileNet was selected as the image recognition model. It is believed that the product will aid zoos and educators in teaching the kids about wildlife.
Actions (login required)
![[img]](http://eprints.utar.edu.my/style/images/fileicons/application_pdf.png)
